Tiny toolkit for training and sampling a Poké-themed Stable Diffusion LoRA.
train.py– optional LoRA training entry point.generate_some.py– geneate some creature based on predefined prompts.inference.py– single prompt.sanity_check.py– quick script to verify that the LoRA was properly loaded.
- Python 3.10+ and (ideally) a CUDA GPU.
- A LoRA exported into
pokeforge_lora/unet_loraandpokeforge_lora/text_encoder_lora. - Dependencies:
python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt
python train.pyYou can customize training parameters directly in train.py like LEARNING_RATE, BATCH_SIZE, NUM_EPOCHS, etc.
python inference.pyYou can customize the prompt directly in inference.py.
python generate_some.py- You can modify and customize the prompts on the
PROMPTSlist ingenerate_some.py.
A few samples produced with the script (see samples_pokeforge/):
